Commit graph

14 commits

Author SHA1 Message Date
99ca24dbbc Merge upstream origin/work 5f99533 into nigig fork alongside dev 2026-09-25 17:32:26 +03:00
a0f46b7d83 Merge upstream origin/dev fda40f56 into nigig fork - preserve nigig hub/web_server, take upstream dev updates 2026-09-25 17:25:35 +03:00
Admin
0f39445982 ai: Claude Desktop drives any app with the F10 panel -- pick "Claude Desktop" in the panel's provider menu and the app serves its tools to it over MCP
- The loopback MCP server moves from Director into libs/ai/services (mcp::server); Director re-exports it, its lane tokens and tests unchanged. The dispatcher now names the server and its instructions; a TokenStore can live in memory only (ephemeral).
- mcp::host: while Claude Desktop is the provider the panel serves its registry's tools (service__tool names) on an ephemeral 127.0.0.1 port with a fresh bearer token, and writes ~/.makepad/mcp/<exe-stem>.json {pid, port, token, title} (0600 in a 0700 dir), removed when the provider changes or the panel goes. Calls queue to the UI thread and run through EngineCore::call_external: a card in the transcript, destructive calls held for the person's confirm (the pane opens for it), the result sent back when the card lands.
- mcp::mcpb: "Connect to Claude Desktop" writes <exe-stem>.mcpb (a stored zip, manifest_version 0.3, binary server = this executable with --mcp) and opens it so Claude Desktop shows its install dialog.
- platform mcp_relay: `<app> --mcp`, checked first in app_main before any Cx, window, GPU or audio, relays newline JSON-RPC on stdio to the running app's endpoint. It answers initialize/ping itself and tools/list from the app's last list while the app is down, starts the app (detached, MAKEPAD_AI_PROVIDER=claude-desktop, engine up with the pane closed) on the first call that needs it, waits up to 20 s for its file, and exits when stdin closes.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-25 00:43:05 +02:00
Admin
27ae3e2fad director: coding agents ask before acting unless you choose otherwise -- the hard-coded --dangerously-skip-permissions / --dangerously-bypass-approvals-and-sandbox flags are gone; Settings > Coding agents has, per agent (Claude, Codex, Grok), a "Let this agent act without asking (bypass permissions)" checkbox (off by default; adds the provider's own flag, Grok --always-approve) and a custom-arguments field (empty by default; split like shell words without a shell), stored in agent_arguments.ron beside settings.ron and read at every launch and resume.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-24 14:20:07 +02:00
Admin
41fc2e3b0a builder: the Builder TUI matches the new terminal design, Windows executables carry app icons, and every app has one -- plain full-screen menu (SETUP / YOUR LICENSES / MAKEPAD EXPERIMENTS / CODING AGENTS) with one status line and one progress bar; log in with an email kept in the install folder (downloads are no longer personalized); consent screens list each license agreement (Return opens it) with Agree to all / Cancel; a GPU notice screen on Windows and Linux; optional CUDA for Makepad Amp on NVIDIA GPUs; Apple developer tools screen on macOS; updates save local edits as changes/<app>-<date>.diff for a coding agent to merge back; "clean build" runs cargo clean. libs/win_resource writes Windows icon and version resources without rc.exe from the same icon source as macOS bundles; cargo-makepad uses it. Apps get icon.svg + icon_1024.png (tools/app_icons/render.sh). makepad-mac.sh and makepad-windows-preview.sh are the simulated visual specs.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-24 13:48:33 +02:00
Admin
2a3db09653 director: agent lanes form a tree and Grok joins the usage bar -- lanes start child lanes the person can see, Grok's weekly allowance comes from its own billing call, and lanes are hosted by the agents binary on macOS and Linux too
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-09-23 22:34:45 +02:00
Admin
90c9a7b01a tests: the three binaries over ten seconds in release come under it -- the user's law is the whole suite in about two minutes, in release, and the CI now runs every test binary of its selection several at a time with a cap on each, so one long binary is the run's floor. Grok measured every binary the debug-era table named (release harness time): fourteen were already under 10 s and are untouched; three were not. Widgets --lib (1822 tests) 17.4 s -> 8.0 s: no single slow test; every test built a fresh Cx and registered the widget library on it (45 ms each, in release), and libtest runs each test on its own thread, so nothing could be kept. Test modules that build widgets now run their cases through on_test_cx on four threads that stay up and keep one registered Cx each (checkout_test_cx); finger, key, pass, draw-list and window state is cleared on every checkout so a case that locks a sweep or draws a menu leaves nothing for the next. The style-reload tests keep their own contexts: a reload would change the next case. The isolate cost table measures batches of 1 and 2 (the four sizes measured 30.06, 29.92, 29.63 and 29.49 ms per isolate: flat), and the two sheet-contrast tests read one walk of every sheet from a OnceLock instead of building the library twelve times each. Director mcp_battery 10.2 s -> 0.4 s: the slowloris test waited out the production head deadline of 10 s; McpServer::start_with_head_deadline takes the budget, production start still passes HEAD_DEADLINE_MS (asserted), and the test proves the socket is held until an 80 ms budget and closed after it. Piano acoustic_reference 11.6 s -> 8.6 s: the promotion renders are 2.5 s instead of 4.0 s, which covers every sample measure and onset read (2.0 s after an onset found in the first 0.5 s). No test is ignored, deleted or loosened. The review dropped the lane's short-circuit in script_mod (a library change for a test's sake: the one test helper that re-registered on a pooled context no longer does) and its per-module pool thread-locals that the checkout never read.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-22 12:29:54 +02:00
Admin
ab64a64e7e tools: makepad-screen is makepad-agents, and its binary is agents -- the crate is the agent session manager (start, attach, list, the TUI); "screen" was a nod to GNU screen, and its other job, the pty trampoline a terminal starts its shell through, does not need the old name either. The package is makepad-agents in tools/agents, the executable is agents, so it is the command on the PATH itself and the shell wrapper that existed only to give it that name is gone (the binary already resolves the workspace's session directory the wrapper used to export); the terminal's pty spawn looks for agents beside the app, Director looks for and pins the sibling agents with the makepad-agents-v1 record and the agents version line, the session environment is MAKEPAD_AGENTS_SESSION and MAKEPAD_AGENTS_STATE_DIR, the CI scripts build makepad-agents, and Director's notes say how to build and run it
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-22 09:36:13 +02:00
Admin
c091430d32 ci: a quiet header, a build that visibly moves, and scripts that leave the machine alone -- the window's top is one band: the branch and its tip, the test progress large (17 / 29) with failed and warning counts only when there are any, the run's elapsed time, Run now and Stop; the line under it is the run progress bar and the tiles start right below it, the rows of counts, "now:" and "Watching" are gone and next poll and the model's state moved to the one footer line; a running step carries what its command is doing ("compiled makepad_draw · 212 crates", told at most twice a second from cargo's artifact messages) and the running tile shows it, so a ten-minute warm build no longer looks stuck; director embeds a terminal, so its script builds the pty helper as the terminal's does; ci.launch takes app_env, and Files runs on its synthetic tree (MAKEPAD_FILES_DEMO) instead of walking the real home, which made macOS ask the person at the CI box for access to Downloads on behalf of "release"; and the workspace tests run with --no-fail-fast, so a run names every failing test target instead of the first
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 23:41:02 +02:00
Admin
5165eaf349 ci: the terminal gets its pty helper, director is a desktop tool, and a refused input is asked again -- on macOS the terminal starts its shell through makepad-screen, which lives next to it and which an app build alone does not produce, so the terminal and window manager scripts build it (the CI box had a terminal that could not spawn a shell and a desk whose terminal window stayed blank); director drives cargo builds and child processes and most of it is compiled out on web and mobile, so its script asks for the desktop targets only; and when an app answers that it could not place an input on a frame and says retry, the harness asks again, up to five times, instead of failing the step
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 22:44:50 +02:00
Admin
229cdd4d2e ci: tools/ci, a Splash runtime that watches a branch and runs every ci.splash it finds -- each script checks its app on every other platform, builds and runs it here, drives it over --remote and has a small local vision model judge the grabs
A test is a ci.splash beside what it tests; the script decides every input and the model only ever judges one picture against one acceptance text. mod.ci: launch (hidden, --remote, user_seq preserved), key, type_text, click, get, snap, wait_log (a * is a gap inside one line), no_errors, grab, quit; step, sleep, check, run; cargo, check_targets (the cargo makepad check matrix, check only for platforms we are not on, a test fails if the two tables drift), test, build, machine (another box over the makepad tunnel), exclusive; judge, accept, ask. The watcher polls git ls-remote once a minute for work and any extra branches, syncs a checkout the CI owns, runs the root script first and alone, then the rest up to a parallel limit behind one shared model judge. The window is a wall of squares, one per script: green passed, orange warnings, red failures, with a detail panel for the selected one.

Scripts: the root ci.splash (workspace check with core warnings denied, the tests), apps/wm (desktop up, switch to macOS by Cmd+Space / type / Return, launch the terminal and the browser, each waited for by the WM's own first-frame line), and one per main app in the default shape. Proven here: apps/wm/ci.splash green in 280 s, fifteen target checks and seven vision verdicts.

Models come from Hugging Face through the hub: registry entries qwen3.5-4b-vision and qwen3.5-9b-vision with exact revisions, sizes and digests, and hub-install, a command line over LocalModels::start_install. vlm-probe reads PNG.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-21 21:23:11 +02:00
Admin
8315f089e0 platform, wm, director: a hosted app's software frame arrives as a top-left texture and the window manager and the director carry no flip term at all
The child draws its frame through a texture pass, which on GL renders through an inverted projection, so its glReadPixels rows come in picture order already; the host keeps them as they are, and the two run views sample the texture as stored on every path. The old shader flip on the software path inverted it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-20 23:38:22 +02:00
Kevin Boos
5e9a697940
app_main!: only ship the fonts an app declares (#1247)
Every manifest `app_main!` emitted carried three fonts no theme role uses:
`NewCMMath-Regular.otf` for `MathView`, and `Inter.ttf` / `RobotoFlex.ttf` for the opt-in
iOS and Android platform styles. That was 3.6 MB in every package, and it only existed
because a font an app forgot to declare failed silently: `FontFamily::update_font_definitions`
skipped a member whose bytes never arrived, so the text showed as boxes with no log line.

* Drop the forced extras. The manifest is now the font set's fallback chain plus whatever
  the app puts in `font_assets`, which is what makepad's own apps already did for `Inter`.
* `font_assets` takes expressions, and `INTER_FONT_ASSET` / `ROBOTO_FLEX_FONT_ASSET` join
  `MATH_VIEW_FONT_ASSET`, so an app declares a font by name instead of by path.
* A font that never loads now logs one `error!` naming the path and the fix. A missing
  member still degrades gracefully: the family keeps its remaining members.
* The apps that use those fonts declare them: the `wm` family binds both platform faces,
  `clock`, `weather` and `director` draw with Inter and can select any style, `terminal`
  and the builder use Inter for symbols, `splash` and `aichat` use `MathView`.
2026-09-19 20:35:58 +02:00
Admin
3811545d48 apps: director, studio, scope, aichat
Squash of 8 work commits (Sep 2–10):
  7b9ed2c  aichat: the assistant as an app — the panel owns the engine, the bus client, settings with the local-only lock
  e0c6e74  aichat: the progress bar and system lines use the theme's highlight colour
  8b46ce0  aichat: the composer's hint is a dark grey Ask AI, not the typed colour
  b99a631  toml_parser, rust_tokenizer: rewrite both for the code analyser
  3bbcea2  aichat: add Studio evaluation-feedback widget
  b61845f  studio: Architecture view, the third workspace mode
  d7a76cf  studio: add bounded code context and source APIs
  524142a  Split Studio into makepad director (public) and makepad scope (private)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
2026-09-15 13:40:29 +02:00